Google and Yale Develop 27B-Parameter AI for Single-Cell Cancer Research

A collaboration between Google and Yale University has produced a large-scale AI model called Cell2Sentence-Scale 27B (C2S-Scale 27B) designed for detailed single-cell analysis. The model contains 27 billion parameters and builds on Google’s Gemma architecture, expanding its capabilities to biological and cellular research.

The research team demonstrated that C2S-Scale 27B can perform predictive and generative tasks related to cellular behavior, similar to conversational AI systems. The model was developed following earlier findings that larger model sizes improve performance in single-cell analysis tasks. With this release, the team has made the model publicly accessible for research use.

To test its effectiveness, scientists used C2S-Scale 27B to simulate the effects of more than 4,000 drugs. The model predicted that silmitasertib (CX-4945) would enhance immune signaling in the presence of low doses of interferon, a condition relevant to cancer detection. Laboratory experiments confirmed this prediction, showing synergistic effects in antigen presentation when interferon was present.

The research group plans to continue exploring how C2S-Scale 27B can be applied to other immunological and therapeutic studies, including cancer immunotherapy development.